Building AI That Learns and Adapts: A Case Study in MRI Diagnostics
summary
201 MRI MRI series (plus 100 normals and 50 other pathologies) train a U-Net style segmentation pipeline; initial model detects about 70% of FCDs but produces 86 false positives in a prospective 100‑patient run versus 5 from non‑expert doctors. Preprocessing uses tissue segmentation, junction/extension and T1/T2/FLAIR inputs. Rebalancing to ~10% FCD and applying Elastic Weight Consolidation as a continual‑learning strategy yields similar sensitivity while significantly improving specificity and robustness to distribution shifts compared to naive full retraining.
AI-written summary · about
chapters (10)
- 0:00Introduction and use case
- 2:21Detecting FCD with AI
- 9:36Training and evaluation
- 13:42Clinical validation findings
- 18:42Continual learning basics
- 24:53Continual learning methods
- 28:59Applying CL to FCD detection
- 35:37Practical applications
- 37:23Future of continual learning
- 42:14Conclusion
related talks
49mBuilding AI That Learns and Adapts: A Case Study in MRI Diagnostics
50mBuilding AI That Learns and Adapts: A Case Study in MRI Diagnostics
42mBuilding AI That Learns and Adapts: A Case Study in MRI Diagnostics
44mBuilding Adaptive AI for MRI Diagnostics: A Continual Learning Case Study
15mBrain Tumor Detection and Classification Using Deep Learning Models
34mWhy AI in the Medical Field Both Sucks and Rocks...
47mThe whole truth about using deep learning in medical imaging
55mDeep Learning in Medical Image Diagnostics